Add additional metrics for container memory

5 files changed, 35 insertions, 8 deletions

diff --git a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/cgroup/MemoryController.java b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/cgroup/MemoryController.java
index 840cd025917..3dc32e14a9d 100644
--- a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/cgroup/MemoryController.java
+++ b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/cgroup/MemoryController.java
@@ -31,9 +31,11 @@ public class MemoryController {
         return cgroup.readIfExists("memory.current").map(Size::from);
     }
 
-    /** @return Number of bytes used to cache filesystem data, including tmpfs and shared memory. */
-    public Size readFileSystemCache() {
-        return Size.from(readField(cgroup.readLines("memory.stat"), "file"));
+    public Stats readStat() {
+        var lines = cgroup.readLines("memory.stat");
+        return new Stats(
+                Size.from(readField(lines, "file")), Size.from(readField(lines, "sock")), Size.from(readField(lines, "slab")),
+                Size.from(readField(lines, "slab_reclaimable")), Size.from(readField(lines, "anon")));
     }
 
     private static String readField(List<String> lines, String fieldName) {
@@ -45,4 +47,13 @@ public class MemoryController {
                     .findFirst()
                     .orElseThrow(() -> new IllegalArgumentException("No such field: " + fieldName));
     }
+
+    /**
+     * @param file Number of bytes used to cache filesystem data, including tmpfs and shared memory.
+     * @param sock Amount of memory used in network transmission buffers.
+     * @param slab Amount of memory used for storing in-kernel data structures.
+     * @param slabReclaimable Part of "slab" that might be reclaimed, such as dentries and inodes.
+     * @param anon Amount of memory used in anonymous mappings such as brk(), sbrk(), and mmap(MAP_ANONYMOUS).
+     */
+    public record Stats(Size file, Size sock, Size slab, Size slabReclaimable, Size anon) {}
 }
diff --git a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/container/ContainerStats.java b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/container/ContainerStats.java
index a5606784c12..1c02072ed2b 100644
--- a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/container/ContainerStats.java
+++ b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/container/ContainerStats.java
@@ -44,8 +44,14 @@ public record ContainerStats(Map<String, NetworkStats> networks,
      * @param cache memory used by cache in bytes
      * @param usage memory usage in bytes
      * @param limit memory limit in bytes
+     * @param sock network transmission buffers in bytes
+     * @param slab in-kernel data structures in bytes
+     * @param slabReclaimable part of "slab" that might be reclaimed in bytes
+     * @param anon anonymous mappings in bytes
      */
-    public record MemoryStats(long cache, long usage, long limit) {}
+    public record MemoryStats(long cache, long usage, long limit, long sock, long slab, long slabReclaimable, long anon) {
+        public MemoryStats(long cache, long usage, long limit) { this(cache, usage, limit, 0, 0, 0, 0); }
+    }
 
     /**
      * Statistics for CPU usage
diff --git a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/container/ContainerStatsCollector.java b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/container/ContainerStatsCollector.java
index 8244666f9e0..0e16e2cabf6 100644
--- a/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/container/ContainerStatsCollector.java
+++ b/node-admin/src/main/java/com/yahoo/vespa/hosted/node/admin/container/ContainerStatsCollector.java
@@ -105,8 +105,10 @@ class ContainerStatsCollector {
         MemoryController memoryController = rootCgroup.resolveContainer(containerId).memory();
         Size max = memoryController.readMax();
         long memoryUsageInBytes = memoryController.readCurrent().value();
-        long cachedInBytes = memoryController.readFileSystemCache().value();
-        return new ContainerStats.MemoryStats(cachedInBytes, memoryUsageInBytes, max.isMax() ? -1 : max.value());
+        var stats = memoryController.readStat();
+        return new ContainerStats.MemoryStats(
+                stats.file().value(), memoryUsageInBytes, max.isMax() ? -1 : max.value(),
+                stats.sock().value(), stats.slab().value(), stats.slabReclaimable().value(), stats.anon().value());
     }
 
     private ContainerStats.NetworkStats collectNetworkStats(String iface, int containerPid) throws IOException {
diff --git a/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/cgroup/CgroupTest.java b/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/cgroup/CgroupTest.java
index c93d90329f7..dd81ea8e76a 100644
--- a/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/cgroup/CgroupTest.java
+++ b/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/cgroup/CgroupTest.java
@@ -106,8 +106,14 @@ public class CgroupTest {
                 shmem 8380416
                 file_mapped 1081344
                 file_dirty 135168
+                slab_reclaimable 1424320
                 """);
-        assertEquals(69206016L, containerCgroup.memory().readFileSystemCache().value());
+        var stats = containerCgroup.memory().readStat();
+        assertEquals(69206016L, stats.file().value());
+        assertEquals(3481600L, stats.anon().value());
+        assertEquals(3552304L, stats.slab().value());
+        assertEquals(73728L, stats.sock().value());
+        assertEquals(1424320L, stats.slabReclaimable().value());
     }
 
     @Test
diff --git a/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/container/ContainerStatsCollectorTest.java b/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/container/ContainerStatsCollectorTest.java
index d4598c8923f..2990e881640 100644
--- a/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/container/ContainerStatsCollectorTest.java
+++ b/node-admin/src/test/java/com/yahoo/vespa/hosted/node/admin/container/ContainerStatsCollectorTest.java
@@ -2,6 +2,7 @@
 package com.yahoo.vespa.hosted.node.admin.container;
 
 import com.yahoo.vespa.hosted.node.admin.cgroup.Cgroup;
+import com.yahoo.vespa.hosted.node.admin.cgroup.MemoryController;
 import com.yahoo.vespa.hosted.node.admin.cgroup.Size;
 import com.yahoo.vespa.hosted.node.admin.configserver.noderepository.NodeSpec;
 import com.yahoo.vespa.hosted.node.admin.nodeagent.NodeAgentContext;
@@ -97,7 +98,8 @@ public class ContainerStatsCollectorTest {
     private void mockMemoryStats(ContainerId containerId) {
         when(cgroup.resolveContainer(eq(containerId)).memory().readCurrent()).thenReturn(Size.from(1228017664L));
         when(cgroup.resolveContainer(eq(containerId)).memory().readMax()).thenReturn(Size.from(2147483648L));
-        when(cgroup.resolveContainer(eq(containerId)).memory().readFileSystemCache()).thenReturn(Size.from(470790144L));
+        when(cgroup.resolveContainer(eq(containerId)).memory().readStat()).thenReturn(
+                new MemoryController.Stats(Size.from(470790144L), Size.from(0), Size.from(0), Size.from(0), Size.from(0)));
     }
 
     private void mockCpuStats(ContainerId containerId) throws IOException {